Abstract:
This project presents BShotT, an automated cricket batting technique analysis system that uses video-based motion tracking, pose estimation, and deep learning to deliver real-time, data-driven feedback to players and coaches. Traditional coaching often relies heavily on subjective observation, making it difficult for athletes to identify technical errors immediately. BShotT addresses this gap by analysing batting movements through MediaPipe pose estimation, extracting key joint angles—such as elbow, shoulder, knee, body, and facial alignment—from each video frame. These angle sequences are processed using a Long Short-Term Memory (LSTM) model to classify shots into common categories, including Cover Drive, Pull Shot, Cut Shot, Sweep Shot and Defence.
The system supports both live webcam input and offline video uploads, enabling flexible use across various training environments. Real-time results are generated through a Flask backend that processes frames via OpenCV, performs model inference, and returns detailed joint-level feedback to a React.js front-end. Players receive moment-by-moment posture corrections, ideal-angle comparisons, and accuracy scoring, allowing them to refine technique without requiring continuous coach supervision.
Testing demonstrated high accuracy in shot classification, low-latency feedback, stable real-time performance, and consistent pose tracking under standard conditions. While challenges such as lighting variation and limited dataset diversity affected some scenarios, the system proved effective as an intelligent training assistant. BShotT highlights the potential of integrating machine learning with sports biomechanics, offering a scalable foundation for advanced cricket analytics, personalized coaching, and future enhancements such as expanded datasets, left-handed support, mobile deployment, and contextual biomechanics analysis.
